ivshmem: remove library and its EAL integration
[dpdk.git] / lib / librte_eal / linuxapp / eal / eal.c
index 6c8e760..d5b81a3 100644 (file)
@@ -1,7 +1,7 @@
 /*-
  *   BSD LICENSE
  *
- *   Copyright(c) 2010-2014 Intel Corporation. All rights reserved.
+ *   Copyright(c) 2010-2016 Intel Corporation. All rights reserved.
  *   Copyright(c) 2012-2014 6WIND S.A.
  *   All rights reserved.
  *
 #include <getopt.h>
 #include <sys/file.h>
 #include <fcntl.h>
-#include <dlfcn.h>
 #include <stddef.h>
 #include <errno.h>
 #include <limits.h>
 #include <errno.h>
 #include <sys/mman.h>
 #include <sys/queue.h>
+#include <sys/stat.h>
+#if defined(RTE_ARCH_X86)
 #include <sys/io.h>
+#endif
 
 #include <rte_common.h>
 #include <rte_debug.h>
 #include <rte_memory.h>
 #include <rte_memzone.h>
 #include <rte_launch.h>
-#include <rte_tailq.h>
 #include <rte_eal.h>
 #include <rte_eal_memconfig.h>
 #include <rte_per_lcore.h>
@@ -74,7 +75,6 @@
 #include <rte_version.h>
 #include <rte_atomic.h>
 #include <malloc_heap.h>
-#include <rte_eth_ring.h>
 
 #include "eal_private.h"
 #include "eal_thread.h"
 #include "eal_filesystem.h"
 #include "eal_hugepages.h"
 #include "eal_options.h"
+#include "eal_vfio.h"
 
 #define MEMSIZE_IF_NO_HUGE_PAGE (64ULL * 1024ULL * 1024ULL)
 
 #define SOCKET_MEM_STRLEN (RTE_MAX_NUMA_NODES * 10)
 
-#define HIGHEST_RPL 3
-
 /* Allow the application to print its usage message too if set */
 static rte_usage_hook_t        rte_application_usage_hook = NULL;
 
-TAILQ_HEAD(shared_driver_list, shared_driver);
-
-/* Definition for shared object drivers. */
-struct shared_driver {
-       TAILQ_ENTRY(shared_driver) next;
-
-       char    name[PATH_MAX];
-       void*   lib_handle;
-};
-
-/* List of external loadable drivers */
-static struct shared_driver_list solib_list =
-TAILQ_HEAD_INITIALIZER(solib_list);
-
 /* early configuration structure, when memory config is not mmapped */
 static struct rte_mem_config early_mem_config;
 
@@ -286,7 +271,7 @@ rte_eal_config_reattach(void)
 }
 
 /* Detect if we are a primary or a secondary process */
-static enum rte_proc_type_t
+enum rte_proc_type_t
 eal_proc_type_detect(void)
 {
        enum rte_proc_type_t ptype = RTE_PROC_PRIMARY;
@@ -309,9 +294,7 @@ eal_proc_type_detect(void)
 static void
 rte_config_init(void)
 {
-       rte_config.process_type = (internal_config.process_type == RTE_PROC_AUTO) ?
-                       eal_proc_type_detect() : /* for auto, detect the type */
-                       internal_config.process_type; /* otherwise use what's already set */
+       rte_config.process_type = internal_config.process_type;
 
        switch (rte_config.process_type){
        case RTE_PROC_PRIMARY:
@@ -354,17 +337,13 @@ eal_usage(const char *prgname)
        printf("\nUsage: %s ", prgname);
        eal_common_usage();
        printf("EAL Linux options:\n"
-              "  -d LIB.so    : add driver (can be used multiple times)\n"
-              "  --"OPT_XEN_DOM0" : support application running on Xen Domain0 "
-                          "without hugetlbfs\n"
-              "  --"OPT_SOCKET_MEM" : memory to allocate on specific\n"
-                  "                 sockets (use comma separated values)\n"
-              "  --"OPT_HUGE_DIR"   : directory where hugetlbfs is mounted\n"
-              "  --"OPT_FILE_PREFIX": prefix for hugepage filenames\n"
-              "  --"OPT_BASE_VIRTADDR": specify base virtual address\n"
-              "  --"OPT_VFIO_INTR": specify desired interrupt mode for VFIO "
-                          "(legacy|msi|msix)\n"
-              "  --"OPT_CREATE_UIO_DEV": create /dev/uioX (usually done by hotplug)\n"
+              "  --"OPT_SOCKET_MEM"        Memory to allocate on sockets (comma separated values)\n"
+              "  --"OPT_HUGE_DIR"          Directory where hugetlbfs is mounted\n"
+              "  --"OPT_FILE_PREFIX"       Prefix for hugepage filenames\n"
+              "  --"OPT_BASE_VIRTADDR"     Base virtual address\n"
+              "  --"OPT_CREATE_UIO_DEV"    Create /dev/uioX (usually done by hotplug)\n"
+              "  --"OPT_VFIO_INTR"         Interrupt mode for VFIO (legacy|msi|msix)\n"
+              "  --"OPT_XEN_DOM0"          Support running on Xen dom0 without hugetlbfs\n"
               "\n");
        /* Allow the application to print its usage message too if hook is set */
        if ( rte_application_usage_hook ) {
@@ -450,14 +429,17 @@ eal_parse_base_virtaddr(const char *arg)
                return -1;
 
        /* make sure we don't exceed 32-bit boundary on 32-bit target */
-#ifndef RTE_ARCH_X86_64
+#ifndef RTE_ARCH_64
        if (addr >= UINTPTR_MAX)
                return -1;
 #endif
 
-       /* align the addr on 2M boundary */
-       internal_config.base_virtaddr = RTE_PTR_ALIGN_CEIL((uintptr_t)addr,
-                                                          RTE_PGSIZE_2M);
+       /* align the addr on 16M boundary, 16MB is the minimum huge page
+        * size on IBM Power architecture. If the addr is aligned to 16MB,
+        * it can align to 2MB for x86. So this alignment can also be used
+        * on x86 */
+       internal_config.base_virtaddr =
+               RTE_PTR_ALIGN_CEIL((uintptr_t)addr, (size_t)RTE_PGSIZE_16M);
 
        return 0;
 }
@@ -484,102 +466,85 @@ eal_parse_vfio_intr(const char *mode)
        return -1;
 }
 
-static inline size_t
-eal_get_hugepage_mem_size(void)
+/* Parse the arguments for --log-level only */
+static void
+eal_log_level_parse(int argc, char **argv)
 {
-       uint64_t size = 0;
-       unsigned i, j;
-
-       for (i = 0; i < internal_config.num_hugepage_sizes; i++) {
-               struct hugepage_info *hpi = &internal_config.hugepage_info[i];
-               if (hpi->hugedir != NULL) {
-                       for (j = 0; j < RTE_MAX_NUMA_NODES; j++) {
-                               size += hpi->hugepage_sz * hpi->num_pages[j];
-                       }
-               }
+       int opt;
+       char **argvopt;
+       int option_index;
+       const int old_optind = optind;
+       const int old_optopt = optopt;
+       char * const old_optarg = optarg;
+
+       argvopt = argv;
+       optind = 1;
+
+       eal_reset_internal_config(&internal_config);
+
+       while ((opt = getopt_long(argc, argvopt, eal_short_options,
+                                 eal_long_options, &option_index)) != EOF) {
+
+               int ret;
+
+               /* getopt is not happy, stop right now */
+               if (opt == '?')
+                       break;
+
+               ret = (opt == OPT_LOG_LEVEL_NUM) ?
+                       eal_parse_common_option(opt, optarg, &internal_config) : 0;
+
+               /* common parser is not happy */
+               if (ret < 0)
+                       break;
        }
 
-       return (size < SIZE_MAX) ? (size_t)(size) : SIZE_MAX;
+       /* restore getopt lib */
+       optind = old_optind;
+       optopt = old_optopt;
+       optarg = old_optarg;
 }
 
 /* Parse the argument given in the command line of the application */
 static int
 eal_parse_args(int argc, char **argv)
 {
-       int opt, ret, i;
+       int opt, ret;
        char **argvopt;
        int option_index;
-       int coremask_ok = 0;
        char *prgname = argv[0];
-       struct shared_driver *solib;
+       const int old_optind = optind;
+       const int old_optopt = optopt;
+       char * const old_optarg = optarg;
 
        argvopt = argv;
-
-       internal_config.memory = 0;
-       internal_config.force_nrank = 0;
-       internal_config.force_nchannel = 0;
-       internal_config.hugefile_prefix = HUGEFILE_PREFIX_DEFAULT;
-       internal_config.hugepage_dir = NULL;
-       internal_config.force_sockets = 0;
-       internal_config.syslog_facility = LOG_DAEMON;
-       /* default value from build option */
-       internal_config.log_level = RTE_LOG_LEVEL;
-       internal_config.xen_dom0_support = 0;
-       /* if set to NONE, interrupt mode is determined automatically */
-       internal_config.vfio_intr_mode = RTE_INTR_MODE_NONE;
-#ifdef RTE_LIBEAL_USE_HPET
-       internal_config.no_hpet = 0;
-#else
-       internal_config.no_hpet = 1;
-#endif
-       /* zero out the NUMA config */
-       for (i = 0; i < RTE_MAX_NUMA_NODES; i++)
-               internal_config.socket_mem[i] = 0;
-
-       /* zero out hugedir descriptors */
-       for (i = 0; i < MAX_HUGEPAGE_SIZES; i++)
-               internal_config.hugepage_info[i].lock_descriptor = -1;
-
-       internal_config.vmware_tsc_map = 0;
-       internal_config.base_virtaddr = 0;
+       optind = 1;
 
        while ((opt = getopt_long(argc, argvopt, eal_short_options,
                                  eal_long_options, &option_index)) != EOF) {
 
-               int ret;
-
                /* getopt is not happy, stop right now */
-               if (opt == '?')
-                       return -1;
+               if (opt == '?') {
+                       eal_usage(prgname);
+                       ret = -1;
+                       goto out;
+               }
 
                ret = eal_parse_common_option(opt, optarg, &internal_config);
                /* common parser is not happy */
                if (ret < 0) {
                        eal_usage(prgname);
-                       return -1;
+                       ret = -1;
+                       goto out;
                }
                /* common parser handled this option */
-               if (ret == 0) {
-                       /* special case, note that the common parser accepted
-                        * the coremask option */
-                       if (opt == 'c')
-                               coremask_ok = 1;
+               if (ret == 0)
                        continue;
-               }
 
                switch (opt) {
-               /* force loading of external driver */
-               case 'd':
-                       solib = malloc(sizeof(*solib));
-                       if (solib == NULL) {
-                               RTE_LOG(ERR, EAL, "malloc(solib) failed\n");
-                               return -1;
-                       }
-                       memset(solib, 0, sizeof(*solib));
-                       strncpy(solib->name, optarg, PATH_MAX-1);
-                       solib->name[PATH_MAX-1] = 0;
-                       TAILQ_INSERT_TAIL(&solib_list, solib, next);
-                       break;
+               case 'h':
+                       eal_usage(prgname);
+                       exit(EXIT_SUCCESS);
 
                /* long options */
                case OPT_XEN_DOM0_NUM:
@@ -589,7 +554,8 @@ eal_parse_args(int argc, char **argv)
                        RTE_LOG(ERR, EAL, "Can't support DPDK app "
                                "running on Dom0, please configure"
                                " RTE_LIBRTE_XEN_DOM0=y\n");
-                       return -1;
+                       ret = -1;
+                       goto out;
 #endif
                        break;
 
@@ -606,7 +572,8 @@ eal_parse_args(int argc, char **argv)
                                RTE_LOG(ERR, EAL, "invalid parameters for --"
                                                OPT_SOCKET_MEM "\n");
                                eal_usage(prgname);
-                               return -1;
+                               ret = -1;
+                               goto out;
                        }
                        break;
 
@@ -615,7 +582,8 @@ eal_parse_args(int argc, char **argv)
                                RTE_LOG(ERR, EAL, "invalid parameter for --"
                                                OPT_BASE_VIRTADDR "\n");
                                eal_usage(prgname);
-                               return -1;
+                               ret = -1;
+                               goto out;
                        }
                        break;
 
@@ -624,7 +592,8 @@ eal_parse_args(int argc, char **argv)
                                RTE_LOG(ERR, EAL, "invalid parameters for --"
                                                OPT_VFIO_INTR "\n");
                                eal_usage(prgname);
-                               return -1;
+                               ret = -1;
+                               goto out;
                        }
                        break;
 
@@ -646,76 +615,42 @@ eal_parse_args(int argc, char **argv)
                                        "on Linux\n", opt);
                        }
                        eal_usage(prgname);
-                       return -1;
+                       ret = -1;
+                       goto out;
                }
        }
 
-       /* sanity checks */
-       if (!coremask_ok) {
-               RTE_LOG(ERR, EAL, "coremask not specified\n");
-               eal_usage(prgname);
-               return -1;
-       }
-       if (internal_config.process_type == RTE_PROC_AUTO){
-               internal_config.process_type = eal_proc_type_detect();
-       }
-       if (internal_config.process_type == RTE_PROC_INVALID){
-               RTE_LOG(ERR, EAL, "Invalid process type specified\n");
-               eal_usage(prgname);
-               return -1;
-       }
-       if (internal_config.process_type == RTE_PROC_PRIMARY &&
-                       internal_config.force_nchannel == 0) {
-               RTE_LOG(ERR, EAL, "Number of memory channels (-n) not specified\n");
-               eal_usage(prgname);
-               return -1;
+       if (eal_adjust_config(&internal_config) != 0) {
+               ret = -1;
+               goto out;
        }
-       if (index(internal_config.hugefile_prefix,'%') != NULL){
-               RTE_LOG(ERR, EAL, "Invalid char, '%%', in '"OPT_FILE_PREFIX"' option\n");
-               eal_usage(prgname);
-               return -1;
-       }
-       if (internal_config.memory > 0 && internal_config.force_sockets == 1) {
-               RTE_LOG(ERR, EAL, "Options -m and --socket-mem cannot be specified "
-                               "at the same time\n");
+
+       /* sanity checks */
+       if (eal_check_common_options(&internal_config) != 0) {
                eal_usage(prgname);
-               return -1;
-       }
-       /* --no-huge doesn't make sense with either -m or --socket-mem */
-       if (internal_config.no_hugetlbfs &&
-                       (internal_config.memory > 0 ||
-                                       internal_config.force_sockets == 1)) {
-               RTE_LOG(ERR, EAL, "Options -m or --socket-mem cannot be specified "
-                               "together with --no-huge!\n");
-               eal_usage(prgname);
-               return -1;
+               ret = -1;
+               goto out;
        }
+
        /* --xen-dom0 doesn't make sense with --socket-mem */
        if (internal_config.xen_dom0_support && internal_config.force_sockets == 1) {
-               RTE_LOG(ERR, EAL, "Options --socket-mem cannot be specified "
-                                       "together with --xen_dom0!\n");
+               RTE_LOG(ERR, EAL, "Options --"OPT_SOCKET_MEM" cannot be specified "
+                       "together with --"OPT_XEN_DOM0"\n");
                eal_usage(prgname);
-               return -1;
-       }
-
-       if (rte_eal_devargs_type_count(RTE_DEVTYPE_WHITELISTED_PCI) != 0 &&
-               rte_eal_devargs_type_count(RTE_DEVTYPE_BLACKLISTED_PCI) != 0) {
-               RTE_LOG(ERR, EAL, "Error: blacklist [-b] and whitelist "
-                       "[-w] options cannot be used at the same time\n");
-               eal_usage(prgname);
-               return -1;
+               ret = -1;
+               goto out;
        }
 
        if (optind >= 0)
                argv[optind-1] = prgname;
+       ret = optind-1;
 
-       /* if no memory amounts were requested, this will result in 0 and
-        * will be overriden later, right after eal_hugepage_info_init() */
-       for (i = 0; i < RTE_MAX_NUMA_NODES; i++)
-               internal_config.memory += internal_config.socket_mem[i];
+out:
+       /* restore getopt lib */
+       optind = old_optind;
+       optopt = old_optopt;
+       optarg = old_optarg;
 
-       ret = optind-1;
-       optind = 0; /* reset getopt lib */
        return ret;
 }
 
@@ -754,13 +689,46 @@ rte_eal_mcfg_complete(void)
 
 /*
  * Request iopl privilege for all RPL, returns 0 on success
+ * iopl() call is mostly for the i386 architecture. For other architectures,
+ * return -1 to indicate IO privilege can't be changed in this way.
  */
-static int
+int
 rte_eal_iopl_init(void)
 {
-       return iopl(HIGHEST_RPL);
+#if defined(RTE_ARCH_X86)
+       if (iopl(3) != 0)
+               return -1;
+#endif
+       return 0;
 }
 
+#ifdef VFIO_PRESENT
+static int rte_eal_vfio_setup(void)
+{
+       int vfio_enabled = 0;
+
+       if (!internal_config.no_pci) {
+               pci_vfio_enable();
+               vfio_enabled |= pci_vfio_is_enabled();
+       }
+
+       if (vfio_enabled) {
+
+               /* if we are primary process, create a thread to communicate with
+                * secondary processes. the thread will use a socket to wait for
+                * requests from secondary process to send open file descriptors,
+                * because VFIO does not allow multiple open descriptors on a group or
+                * VFIO container.
+                */
+               if (internal_config.process_type == RTE_PROC_PRIMARY &&
+                               vfio_mp_sync_setup() < 0)
+                       return -1;
+       }
+
+       return 0;
+}
+#endif
+
 /* Launch threads, called at application init(). */
 int
 rte_eal_init(int argc, char **argv)
@@ -768,8 +736,9 @@ rte_eal_init(int argc, char **argv)
        int i, fctret, ret;
        pthread_t thread_id;
        static rte_atomic32_t run_once = RTE_ATOMIC32_INIT(0);
-       struct shared_driver *solib = NULL;
        const char *logid;
+       char cpuset[RTE_CPU_AFFINITY_STR_LEN];
+       char thread_name[RTE_MAX_THREAD_NAME_LEN];
 
        if (!rte_atomic32_test_and_set(&run_once))
                return -1;
@@ -782,6 +751,11 @@ rte_eal_init(int argc, char **argv)
        if (rte_eal_log_early_init() < 0)
                rte_panic("Cannot init early logs\n");
 
+       eal_log_level_parse(argc, argv);
+
+       /* set log level as early as possible */
+       rte_set_log_level(internal_config.log_level);
+
        if (rte_eal_cpu_init() < 0)
                rte_panic("Cannot detect lcores\n");
 
@@ -789,9 +763,6 @@ rte_eal_init(int argc, char **argv)
        if (fctret < 0)
                exit(1);
 
-       /* set log level as early as possible */
-       rte_set_log_level(internal_config.log_level);
-
        if (internal_config.no_hugetlbfs == 0 &&
                        internal_config.process_type != RTE_PROC_SECONDARY &&
                        internal_config.xen_dom0_support == 0 &&
@@ -801,8 +772,6 @@ rte_eal_init(int argc, char **argv)
        if (internal_config.memory == 0 && internal_config.force_sockets == 0) {
                if (internal_config.no_hugetlbfs)
                        internal_config.memory = MEMSIZE_IF_NO_HUGE_PAGE;
-               else
-                       internal_config.memory = eal_get_hugepage_mem_size();
        }
 
        if (internal_config.vmware_tsc_map == 1) {
@@ -820,15 +789,12 @@ rte_eal_init(int argc, char **argv)
 
        rte_config_init();
 
-       if (rte_eal_iopl_init() == 0)
-               rte_config.flags |= EAL_FLG_HIGH_IOPL;
-
        if (rte_eal_pci_init() < 0)
                rte_panic("Cannot init PCI\n");
 
-#ifdef RTE_LIBRTE_IVSHMEM
-       if (rte_eal_ivshmem_init() < 0)
-               rte_panic("Cannot init IVSHMEM\n");
+#ifdef VFIO_PRESENT
+       if (rte_eal_vfio_setup() < 0)
+               rte_panic("Cannot init VFIO\n");
 #endif
 
        if (rte_eal_memory_init() < 0)
@@ -843,42 +809,34 @@ rte_eal_init(int argc, char **argv)
        if (rte_eal_tailqs_init() < 0)
                rte_panic("Cannot init tail queues for objects\n");
 
-#ifdef RTE_LIBRTE_IVSHMEM
-       if (rte_eal_ivshmem_obj_init() < 0)
-               rte_panic("Cannot init IVSHMEM objects\n");
-#endif
-
        if (rte_eal_log_init(logid, internal_config.syslog_facility) < 0)
                rte_panic("Cannot init logs\n");
 
        if (rte_eal_alarm_init() < 0)
                rte_panic("Cannot init interrupt-handling thread\n");
 
-       if (rte_eal_intr_init() < 0)
-               rte_panic("Cannot init interrupt-handling thread\n");
-
        if (rte_eal_timer_init() < 0)
                rte_panic("Cannot init HPET or TSC timers\n");
 
        eal_check_mem_on_local_socket();
 
-       rte_eal_mcfg_complete();
-
-       TAILQ_FOREACH(solib, &solib_list, next) {
-               RTE_LOG(INFO, EAL, "open shared lib %s\n", solib->name);
-               solib->lib_handle = dlopen(solib->name, RTLD_NOW);
-               if (solib->lib_handle == NULL)
-                       RTE_LOG(WARNING, EAL, "%s\n", dlerror());
-       }
+       if (eal_plugins_init() < 0)
+               rte_panic("Cannot init plugins\n");
 
        eal_thread_init_master(rte_config.master_lcore);
 
-       RTE_LOG(DEBUG, EAL, "Master core %u is ready (tid=%x)\n",
-               rte_config.master_lcore, (int)thread_id);
+       ret = eal_thread_dump_affinity(cpuset, RTE_CPU_AFFINITY_STR_LEN);
+
+       RTE_LOG(DEBUG, EAL, "Master lcore %u is ready (tid=%x;cpuset=[%s%s])\n",
+               rte_config.master_lcore, (int)thread_id, cpuset,
+               ret == 0 ? "" : "...");
 
        if (rte_eal_dev_init() < 0)
                rte_panic("Cannot init pmd devices\n");
 
+       if (rte_eal_intr_init() < 0)
+               rte_panic("Cannot init interrupt-handling thread\n");
+
        RTE_LCORE_FOREACH_SLAVE(i) {
 
                /*
@@ -897,6 +855,15 @@ rte_eal_init(int argc, char **argv)
                                     eal_thread_loop, NULL);
                if (ret != 0)
                        rte_panic("Cannot create thread\n");
+
+               /* Set thread_name for aid in debugging. */
+               snprintf(thread_name, RTE_MAX_THREAD_NAME_LEN,
+                       "lcore-slave-%d", i);
+               ret = rte_thread_setname(lcore_config[i].thread_id,
+                                               thread_name);
+               if (ret != 0)
+                       RTE_LOG(DEBUG, EAL,
+                               "Cannot set name for lcore thread\n");
        }
 
        /*
@@ -910,6 +877,8 @@ rte_eal_init(int argc, char **argv)
        if (rte_eal_pci_probe())
                rte_panic("Cannot probe PCI\n");
 
+       rte_eal_mcfg_complete();
+
        return fctret;
 }
 
@@ -917,16 +886,50 @@ rte_eal_init(int argc, char **argv)
 enum rte_lcore_role_t
 rte_eal_lcore_role(unsigned lcore_id)
 {
-       return (rte_config.lcore_role[lcore_id]);
+       return rte_config.lcore_role[lcore_id];
 }
 
 enum rte_proc_type_t
 rte_eal_process_type(void)
 {
-       return (rte_config.process_type);
+       return rte_config.process_type;
 }
 
 int rte_eal_has_hugepages(void)
 {
        return ! internal_config.no_hugetlbfs;
 }
+
+int
+rte_eal_check_module(const char *module_name)
+{
+       char sysfs_mod_name[PATH_MAX];
+       struct stat st;
+       int n;
+
+       if (NULL == module_name)
+               return -1;
+
+       /* Check if there is sysfs mounted */
+       if (stat("/sys/module", &st) != 0) {
+               RTE_LOG(DEBUG, EAL, "sysfs is not mounted! error %i (%s)\n",
+                       errno, strerror(errno));
+               return -1;
+       }
+
+       /* A module might be built-in, therefore try sysfs */
+       n = snprintf(sysfs_mod_name, PATH_MAX, "/sys/module/%s", module_name);
+       if (n < 0 || n > PATH_MAX) {
+               RTE_LOG(DEBUG, EAL, "Could not format module path\n");
+               return -1;
+       }
+
+       if (stat(sysfs_mod_name, &st) != 0) {
+               RTE_LOG(DEBUG, EAL, "Module %s not found! error %i (%s)\n",
+                       sysfs_mod_name, errno, strerror(errno));
+               return 0;
+       }
+
+       /* Module has been found */
+       return 1;
+}